Items Based Fuzzy C-Mean Clustering for Collaborative Filtering

Abstract

Collaborative Filtering is a method behind the successful of Recommendation System that is widely used especially in E-Commerce website. It boosts profit for E-Commerce website by trying to predict user interested from peer’s opinions and offering proper products that match user’s interested. A challenge for collaborative filtering is data characteristics i.e. amount of data always large and contains a lot of missing values. Hence it is not easy to perform high recommendation accuracy. In addition, the computational cost of Collaborative Filtering is always highly expensive from scalability issue and hard to predict from cold start and sparsity. There are several approaches but Item based approach is effcient and easy algorithm that reduces effect of scalability by performing calculation on item side instead of user side. In this paper we propose an approach to improve Item based method by applying Fuzzy C-Mean algorithm over Item based to partition items into several clusters and perform prediction against clusters. Primary advantage is greatly reduces computation cost on MovieLens dataset. Our approach shows that it
overcomes scalability, cold start and sparsity. It saves more than 99% computational time and does not change the prediction quality eventually real-time prediction and website responsive can be processed much faster.